Laufwerk documentation
From a small checked result to a workflow of your own.
Laufwerk is a TypeScript framework for multi-step work with AI agents. You define the process, checks and human decisions. Laufwerk records runs and executes agent work in isolated environments on your host.
Start with a result
Fix a small calculation, run its tests, and inspect the verified change. Then adapt the workflow. You need a supported Ubuntu host, Bun and a Codex subscription login. Prepare your host if these are new to you.
Learn as you build
| Your next question | Read |
|---|---|
| What is a workflow, session or workspace? | How it works |
| How do I turn my process into code? | Build a workflow |
| Can a person decide before it continues? | Human decisions |
| What else can I build? | Examples |
| Why did it stop or return an old result? | Inspect and debug |
| What are the exact commands? | CLI reference |
These docs cover 0.0.1-alpha.6. Laufwerk is an early local runtime: see current capabilities and limits. No hosted account is needed to read the docs or install Laufwerk. Model execution uses your own provider subscription.